topic Re: Statistical significance of time trends in Statistical Procedures

Statistical significance of time trends

kewong — Fri, 12 Apr 2013 17:44:55 GMT

Hi SAS community,

I was wondering if you guys can help me out with the correct approach in determining whether or not a time trend is statistically significant over time.

Lets say in year 2000, we had 100 cases with over population of 50,000 = 200/100,000 persons/year.

in year 2005, we had 75 cases with populatino of 60,000 = 125/100,000 persons/year.

in year 2010, we had 50 cases over a population of 75,000 = 66.67/100,000 persons/year.

How do I get the p value and determine whether it's statistically different?

Thanks in advance for all your help!

Re: Statistical significance of time trends

PGStats — Fri, 12 Apr 2013 18:39:15 GMT

You can test for a linear time trend (Cochran-Armitage Test for Trend) as follows :

data test(keep=year case n);
input year cases population;
case = "YES";
n = cases; output;
case = "NO";
n = population-cases; output;
datalines;
2000 100 50000
2005 75 60000
2010 50 75000
;

proc freq data=test;
weight n;
tables year*case / trend;
run;

Re: Statistical significance of time trends

PGStats — Fri, 12 Apr 2013 19:49:39 GMT

Note : The numbers used for testing should reflect actual observations, not population estimates. - PG

Re: Statistical significance of time trends

1zmm — Sat, 13 Apr 2013 18:58:46 GMT

Another possibility is Poisson regression using PROC GENMOD,

data test(keep=year case log_n);

infile datalines;

input year case n;

log_n=log(n);

output test;
datalines;

2000 100 50000

2005 75 60000

2010 50 75000

;

run;

proc genmod data=test;

model case=year / dist=poisson link=log offset=log_n;

run;